[llvm] [LV] Ignore some costs when loop gets fully unrolled (PR #106699)

[Nick Desaulniers via llvm-commits]

================
@@ -2682,6 +2682,25 @@ static Value *getExpandedStep(const InductionDescriptor &ID,
   return I->second;
 }
 
+/// Knowing that loop \p L executes a single vector iteration, add instructions
+/// that will get simplified and thus should not have any cost to \p
+/// InstsToIgnore.
+static void addFullyUnrolledInstructionsToIgnore(
+    Loop *L, const LoopVectorizationLegality::InductionList &IL,
+    SmallPtrSetImpl<Instruction *> &InstsToIgnore) {
+  auto *Cmp = L->getLatchCmpInst();
+  if (Cmp)
+    InstsToIgnore.insert(Cmp);
+  for (const auto &[IV, IndDesc] : IL) {
+    // Get next iteration value of the induction variable.
+    Instruction *IVInst =
+        cast<Instruction>(IV->getIncomingValueForBlock(L->getLoopLatch()));
+    if (all_of(IVInst->users(),
+               [&](const User *U) { return U == IV || U == Cmp; }))
----------------
nickdesaulniers wrote:

https://lab.llvm.org/buildbot/#/builders/78/builds/11135:
```
/home/llvm-libc-buildbot/buildbot-worker/libc-x86_64-debian/libc-x86_64-debian-dbg-runtimes-build/llvm-project/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:2699:49: error: reference to local binding 'IV' declared in enclosing function 'addFullyUnrolledInstructionsToIgnore'
               [&](const User *U) { return U == IV || U == Cmp; }))
                                                ^
/home/llvm-libc-buildbot/buildbot-worker/libc-x86_64-debian/libc-x86_64-debian-dbg-runtimes-build/llvm-project/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:2694:21: note: 'IV' declared here
  for (const auto &[IV, IndDesc] : IL) {
                    ^
```


https://github.com/llvm/llvm-project/pull/106699